More reliable forecasts with less precise computations: a fast-track route to cloud-resolved weather and climate...

Area of Science:

• Atmospheric science
• Computational physics
• Climate modeling

Background:

• Traditional weather and climate models rely on a distinct separation between dynamical cores and sub-grid parametrizations.
• Atmospheric energy spectra exhibit power-law behavior at scales of hundreds of kilometers and below, suggesting inherent stochasticity.

Purpose of the Study:

• To propose a novel methodological approach for weather and climate simulations.
• To develop more energy- and computationally efficient simulation techniques for the exascale computing era.

Main Methods:

• Blurring the conventional boundary between dynamical cores and sub-grid parametrizations.
• Basing closure schemes on stochastic-dynamic systems instead of deterministic formulas.
• Reducing deterministic computation and numerical precision to the largest scales.

Main Results:

• Demonstrates that a stochastic-dynamic approach is more appropriate for weather and climate simulators.
• Highlights the over-engineering of current dynamical cores that use full determinism and precision for all scales.
• Presents an efficient approach for cloud-resolved simulations in the exascale era.

Conclusions:

• A unified, stochastic-dynamic approach is more suitable for simulating atmospheric processes across scales.
• Optimizing computational resources by selectively applying determinism and precision leads to significant efficiency gains.
• This methodology paves the way for more effective and sustainable climate and weather prediction.